🚀 go-pugleaf

I was thinking about upgrading my 386DX/33 with a cheap 486DX/33
motherboard.  The two things I wanted to go faster were games
(had to turn off clouds in Comanche Maximum Overkill) and
JPEG.  So I ran some tests on a friend's Taiwanese 486DX/33
of unknown chipset:

386 - Swan 386DX/33, Micronics MB with either 32 or 64K cache (I'm ashamed 
      to admit I'm not sure which, salesman said 32K, magazine review said
      64K, manual says both are possible), Intel cache controller,  
      8MB ram, Pheonix bios

486 - noname 486DX/33, Taiwanese MB of unknown chipset, 256K external cache,
      16MB ram, AMI bios

Tests:

Tests were run on bare dos 5.0 with minimal stuff in config files:

I used the Independent JPEG Group vesion4 jpeg code, compiled with
the Intel 386/486 C Codebuilder (32bit extended dos) compiler.
The jpeg stuff is not disk bound, and besides, both systems have
comparatively fast IDE drives.  Timing was done with a stopwatch.
lake2.gif is 630x422x256.

cjpeg -Q 80 lake2.gif:

386: 15.0 sec
486: 29.7 sec

djpeg lake2.jpg:

386: 23.3 sec
486: 47.3 sec

The other benchmark results were more reasonable, but these
were canned benchmarks, not real-world tests.

Norton 4.0 cpu index:

386: 42.7
486: 54.2      

Norton 5.0 cpu index:

386: 35.4
486: 72.1

Does anyone have a clue as to why the 486 was twice as slow on the jpeg
stuff?  I know the Intel cache controller on my 386 uses posted writes for 
0ws writes 100% of the time, and the 486's external cache may be 
write-through, but shouldn't the internal writeback cache in the 486 buffer 
the effects of this, if the 486's external cache is indeed write-through.

------------
Bailey Brown                               "Above all else, confusion reigns."
Intergraph Corporation                                  
bbrown@casca.b11.ingr.com                              Procol Harum